14.5.2020

Manuscript

Dynamic fluctuations of emotional states in adolescents with delayed sleep phase—a longitudinal

network modeling approach

Marko Elovainio, PhD 1,2

Liisa Kuula, PhD 1

Risto Halonen, MPsyc 1

Anu-Katriina Pesonen, PhD 1

1 SleepWell Research Program Unit, Department of Psychology and Logopedics, Faculty of

Medicine, University of Helsinki, Finland 2 National Institute for Health and Welfare, Finland

Correspondence: Marko Elovainio, University of Helsinki, P.O. Box 9, 00014, Helsinki, Finland

Phone: +358 50 3020621, email: marko.elovainio@helsinki.fi

Word count for abstract / main text: / 250 / 3531

Number of tables and figures: 1 Table / 3 figures

Funding: Academy of Finland

ABSTRACT

Background: Very late sleep rhythms are risks for social adjustment problems in adolescence.

Using ecological momentary assessment data, we quantified and visualized temporal and

contemporaneous within-persons dynamical relations of sleepiness and emotions in adolescents

with and without late sleep rhythms.

Methods: We analyzed a temporal network via multilevel vector autoregression (mlVAR)

modeling and a contemporaneous network through the partial associations between the residuals of

temporal and the between-subject multilevel models. We tested whether these networks were

different between those with a late circadian rhythm [concurrent delayed sleep phase (DSP) N =

172, 37% boys, 63% girls] and those without (N = 143, 22% boys, 78% girls).

Results: In adolescents without DSP, the temporal networks showed continuity only for low mood

from the previous to the following time point. In adolescents with DSP, there were more predictable

patterns of emotions. Feelings of depression led to a decrease of positive emotions and increase of

irritation and anxiety. The contemporaneous networks showed clusters of positive and negative

emotions in both groups and sleepiness decreased the experience of positive emotions concurrently.

Limitations: DSP in our current study was based only on one out of three diagnostic criteria of the

full disorder (DSM-5) and it was assessed only once.

Conclusions: These findings indicate that the dynamic organization of emotions and sleepiness is

different in adolescents with and without DSP. DSP adolescents have more predictable and

maladaptive emotional patterns during the day. Results provide new insight about why individuals

with DSP are at a heightened risk for decreased emotional adjustment.

INTRODUCTION

Globally, adolescents are sleeping fewer hours than they need (Owens, Adolescent Sleep Working

et al. 2014). This is followed by daytime sleepiness and symptoms of fatigue. Recent surveys

indicate a very high prevalence for hypersomnolence symptoms (>40% in U.S. youth), excessive

daytime sleepiness (>60% in Brazilian youth), or mild daytime sleepiness (>20% in Chinese youth)

during adolescence (Kolla, He et al. 2019, Meyer, Ferrari Junior et al. 2019, Wang, Liu et al. 2019).

Not surprisingly, daytime sleepiness has been associated with poorer sleep and a higher risk of

having mental health problems (Kolla, He et al. 2019).

To better understand adolescent development and the underlying processes, a closer

look at the dynamics of daytime sleepiness and emotional processes is needed. While a body of

evidence shows that quality of sleep and related alertness are associated with negative and positive

emotional states (Caldwell, Caldwell et al. 2004, Kahn-Greene, Killgore et al. 2007, Paterson,

Dorrian et al. 2011, Watling, Pawlik et al. 2017), little is known about their temporal dynamics.

Most of the studies have attempted to examine these associations independently (e.g., the effects of

sleepiness on emotions and the effects of emotions on sleepiness), but such separation does not

reflect the complexity of their mutual dependencies. Moreover, there appears to be a “vicious

cycle” as daytime sleepiness may compromise emotional regulation, which in many cases leads to

increased negative emotion that in turn disrupts sleep, leading to further impairments in alertness

and emotional wellbeing (Mauss, Troy et al. 2013). Understanding these dynamic and complex co-

occurring processes and their underlying mechanisms is of great importance to understand

developmental processes in adolescence, often driven by the common condition of chronically

insufficient sleep (Shochat, Cohen-Zion et al. 2014).

Regarding the dynamics between daytime sleepiness and mood in adolescence, one

group of adolescents is of specific interest: adolescents having a very late sleep rhythm, or delayed

sleep phase (DSP), which is a subclinical form of the delayed sleep phase disorder (DSPD). DSP is

often defined as having some of the three symptoms of the full disorder—most often the

persistently very late sleep rhythm. In population-based studies, this has been operationalized, for

example, as falling asleep after 1 a.m. at least 3 days per week (Danielsson, Markstrom et al. 2016).

DSP in adolescence is often followed by negative consequences in school performance (Gradisar

and Crowley 2013), and it has been associated with depression, anxiety, and symptoms of ADHD

(Sivertsen, Harvey et al. 2015). However, it is unclear how excessive daytime sleepiness and mood

are temporally organized. Tracking the daily fluctuations would give much needed understanding of

the topic.

However, feasible methods to measure and analyze complexity between sleepiness

and mood or other psychological processes have been scarce. To address the temporal dynamics, a

method called ecological momentary assessment (EMA), which uses repeated measurements of

multiple state-like factors with relatively short follow-up intervals, is needed. Regarding EMA and

sleep, previous studies have shown that experiences of fatigue in diseases such as multiple sclerosis

(Kratz, Murphy et al. 2017, Powell, Liossi et al. 2017) vary dynamically across the day, and

emotions are associated with these fluctuations (Powell, Liossi et al. 2017). Previous studies have

also shown that the experiences of psychological problems fluctuate over time, such as suicidal

ideation, even from hour to hour (Hallensleben, Spangenberg et al. 2017), and there are large

intrapersonal differences in these fluctuations.

Recently, a psychological network approach and related tools to analyze dynamic

relationships among multiple psychological variables have been developed (Epskamp, Cramer et al.

2012, Epskamp 2015). The basic idea behind the psychological network approach is that various

psychological states and problems develop as a result of the interaction among individual symptoms

or risk factors (Borsboom 2017). For example, troubles in sleeping may result in concentration

problems, which may lead to more rumination and daytime sleepiness, which in turn may lead to

negative emotional states. When the relations among these risk factors are sufficiently strong, they

can be self-sustained via a negative feedback loop.

Network analysis has so far mostly been applied to cross-sectional data but can also

be applied to EMA data using vector autoregression (VAR) techniques, in which a variable at a

certain time point (t1) is predicted by the same variable at the previous time point (t–1)

(autoregressive effects) and all other variables at t–1 (cross-lagged effects) (Epskamp, Waldorp et

al. 2018). These autoregressive and cross-lagged effects can be quantified and visualized in a

network. The network analyses with VAR techniques have extended more traditional multilevel

analysis by estimating and visualizing the co-occurrence of multiple variables, allowing generation

of hypotheses about the potential dynamic process among multiple variables. By allowing VAR

coefficients to differ across individuals via multilevel modeling, it is possible to model and

visualize temporal dynamics between sleepiness and mood at the group level. Thus, network

analyses could guide researchers and clinicians toward more complex and dynamic thinking about

psychological states (Fried, van Borkulo et al. 2017), such as sleepiness and emotional states. As far

as we know, there are no published studies focusing on the dynamics of emotional states using

longitudinal network approach, although some studies have modelled psychiatric problems,

including suicidality (Rath, de Beurs et al. 2019) and auditory verbal hallucinations (Jongeneel,

Aalbers et al. 2020).”

Network analysis extends standard time series analysis, such as hierarchal linear

modeling, by offering a visual representation of the relations among all assessed variables. In our

study, we applied the VAR method to examine the dynamic associations between daytime

sleepiness and emotional states, including being depressed, anxious, content, irritated, and happy.

To increase understanding of the effects of very late sleep rhythms on sleepiness and emotions, we

compared the sleepiness-emotion networks between those with a concurrent DSP and those without

classified based on actigraphy measurements.

METHODS

Sample and procedure

This research is part of the population-based cohort study SleepHelsinki!. Originally,

the Finnish Population Registry was utilized to identify all Finnish adolescents born between

January 1, 1999, and December 31, 2000 (n=10,476), who resided in Helsinki and whose native

language was registered as Finnish (72% of the total sample). The register thus included 7,539

adolescents (3,789 born in 1999 and 3,750 born in 2000), of whom 50% were boys. We sent

invitation letters to all registered adolescents to participate in the SleepHelsinki! study Phase 1,

which consisted of an online survey primarily targeting sleep, health, and behavior. The estimated

time for filling in the questionnaire was 30 minutes.

Altogether, 1,411 adolescents (19% of the initial cohort) responded to the online

survey, with valid responses from 1,374 (18%) participants. The age of the respondents did not

differ from the initial cohort mean age (p = 0.34), but the respondents were more often girls (66%, p

< 0.0001). All respondents signed an electronic consent form for Phase 1. Ethical permission was

obtained from The Hospital District of Helsinki and Uusimaa Ethics Committee for gynecology and

obstetrics, pediatrics, and psychiatry (Decision number 50/13/03/03/2016). The study is registered

under Clinical Trials (ID: 1287174).

The SleepHelsinki! study (total N = 552) focused on late sleep rhythms. Based by

online survey questionnaire two subsamples: (A) those without (N=188) and (B) those with late

sleep rhythms (N=364) were identified and invited to participate in the EMA phase. The criteria for

belonging to the late sleep phase group was reporting bedtime after 1 a.m. at least three times per

week. Out of these 552 adolescents, 353 agreed to participate and 24 dropped out before completing

the EMA phase, leaving us with a sample of 329 adolescents. The EMA period of the study was

conducted between November 2016 and December 2017. The data for this study were thus from

late adolescence (mean age 16.9 years, SD = 0.6) when the EMA measurements were done. A total

of 315 participants provided usable EMA data and formed the final analyses sample.

The participants underwent a minimum 7-day EMA assessment with three signal-

contingent assessments per day using the “Psymate” software, resulting in a maximum of 66

assessments per participant. The first EMA signal was scheduled in the morning (around 8–10

a.m.), the second in the afternoon (1–5 p.m.), and the final in the evening (around 8–10 p.m.).

Participants could postpone a prompt for 30 minutes if they were not able to answer the questions

immediately, and they had the possibility to reject a prompt. The data set consisted of 4,348

observations (315 persons).

Additionally, adolescents’ sleep timing was measured using accelerometers

(GeneActiv Original, Kimbolton, UK) to determine and confirm possible DSP tendencies. Mean

sleep duration in those with DSP was 7:23 hours (SD=1:01) and in those without 8:09 hours

(SD=0:43). The sleep mid-point in those with DSP was 5:31 (SD=1:05) and in those without 3:50

(SD=0:36).

Measures

During the EMA assessment, participants rated their momentary sleepiness-perceived

depressiveness, anxiousness, irritation, contentedness, and happiness by answering a structured 14-

item questionnaire. At each measurement point, participants reported how alert vs. sleepy they felt

during the last 10 minutes on a 9-point scale with 5 anchors (1 = very alert; 3 = alert; 5 = neither

alert nor sleepy; 7 = sleepy but no trouble staying awake; and 9 = very sleepy, fighting to stay

awake) (Karolinska Sleepiness Scale, KSS).

Additionally, participants reported how they felt using two positive effect and three

negative effect items using 7-point scales from 1 (not at all) to 7 (very much). These specific

emotions were chosen to include those most frequently appearing in several longer affect checklists

(Diener and Emmons 1984, Watson, Clark et al. 1988). The constructs represented were depression,

anxiety (worried), and hostility (irritated) on the negative affect side. Activation (energetic/

sleepiness) and positive feelings (happy, and content) were represented on the positive side. The list

was required to be short to facilitate its use over multiple daily assessments.

The actigraphy measurements covered a minimum of 3 nights (range 3–12 nights) and

yielded a binary classification of concurrent DSP tendency. This information was calculated based

on going to sleep after 1 a.m. at least three times per week. The classification was used to compare

the emotional responses of those with weaker circadian control to those without. Based on the

actigraphy measurements 172 participants were classified as having a late circadian rhythm and

having a concurrent delayed sleep phase (DSP) and 143 were classified as not having DSP (non-

delayed sleep phase / non-DSM) circadian rhythm. Of those with DSM 63% were girls and of those

with non-DSM 78% were girls (X2 = 8.22, df = 1, p-value = 0.004 for sex difference). Parental

education was used as a measure of parental socio-economic status (SES). Participants were asked

to report the highest educational attainment of their parents, and the highest report was selected as

the basis for classifying participants’ SES as (1) lower (2) middle, or, (3) upper. Lower SES

included minimum compulsory educational attainment. Education attained beyond this (i.e.

vocational school, technical college) was classified as middle level SES. SES was classified as

upper level if either parent had university level degree. Most of the parents of those in DSP (N=

119, 73%) and in those non-DSP (N = 105, 77%) belonged to the highest SES category. There were

no differences in parental socioeconomic status measured as education (lower / middle/ upper)

between the DSP and non -DSP group (2(2) = 0.83, p = 0.66).

Statistical analysis

To estimate the different network structures over all participants, we applied

multilevel vector autoregression (mlVAR) models on the data (Epskamp, Waldorp et al. 2018), as

implemented within the mlVAR package in R. In conducting the analyses and presenting the results

we mostly followed the procedure presented by Jongeneel and others (Jongeneel, Aalbers et al.

2020). Temporal dynamics within individuals are estimated by regressing scores of an emotion at

time t on a previous (i.e., lagged) value of itself at t-1. VAR modeling indicates that all variables at

time t are regressed on a t-1 version of themselves, resulting in a vector of lagged regression

coefficients (fixed effects). The multilevel modeling allows the VAR coefficients to differ between

individuals (random effects), and a temporal network was estimated. When visualizing the networks

in a two-dimensional graph, we used the Fruchterman-Reingold (FR) algorithm, which places nodes

that are highly connected in the center of the network. The estimates of between-subject effects, the

sample means within subjects, were used as predictors. The contemporaneous network was

estimated using the residuals of the multilevel model that were used to estimate the temporal and

between-subject effects (Epskamp, Waldorp et al. 2018).

In the graphical representation of the network, variables are presented as nodes. These

networks are called temporal networks because they can be indicative of potential causality given

that one variable preceded the other in time. Epskamp et al. (2018) introduced an additional

network for estimating EMA data (a contemporaneous network) that can be used in the partial

association between the residuals of the temporal network that are the result of associations between

the variables that are not explained by the current chosen time interval, the chosen lag, or anything

else that is not explicitly measured and modeled. In the contemporaneous network, the edges

between nodes represent the partial correlation obtained after controlling for both temporal effects

and all other variables in the same window of measurement (Epskamp, Waldorp, Mõttus, &

Borsboom, 2017; Fisher, Reeves, Lawyer, Medaglia, & Rubel, 2017). When data are collected from

multiple subjects, between-person networks using EMA data may also be estimated. Between-

subject predictors are calculated using the covariance structure of stationary means. In addition, we

calculated the centrality measures of each symptom in each network (strength centrality identifies

how well each emotion in a network is associated with the other emotions). We calculated the

strength of the nodes for all networks, using the R package qgraph (Epskamp et al., 2012). In the

temporal network centrality indicates both how well each emotion is predicted by others (in-

strength) and how well it predicts others (out-strength).

Variables were standardized before estimation and scaled within persons via mlVAR.

Networks were presented using the graph package in R.

RESULTS

Means and standard deviations for all emotions and sleepiness in those with and

without DSP are presented in Table 1 and the distribution of the emotions in the high and low

sleepiness group in Figure 1. Those with DSP more often experienced anxiety, depression, and

irritation, and those without DSP more often were happy and content. There were no differences in

sleepiness reporting between the groups. Only small differences were found in distributions of

emotion reporting between those with and without DSP (Figure 1). Girls were slightly less happy

(mean 4.34 vrs. 4.81, p-value 0.006), less content (mean 4.46 vrs. 5.04, p-value 0.002), and more

anxious (mean 2.32 vrs. 1.88, p-value 0.025) than boys.

Temporal networks

The temporal network including the directed associations is based on time series EMA data that was

analyzed with timelags. All variables at a certain measurement point is predicted by the same

variable and all other variables at the previous measurement point providing information about the

temporal multivariate relations. All coefficients are allowed to differ across individuals by using

multilevel modeling, and thus time dynamics are possible to model and visualize at the group level.

The left top panel in Figure 2 shows how emotions predicted themselves (autoregressions) and each

other across time (adjusted predictions) in those without DSP and the left bottom panel in those

with DSP. Green lines represent positive and red lines (dotted) represent negative associations. The

thickness of the line corresponds to the strength of the association. The panels in the right-hand side

show the centrality measures.

In individuals without DSP, there were no significant patterns observable in the

fluctuation of emotions except a bidirectional loop between depression and irritation (0.12, p =

0.002/0.06, p = 0.17) and a continuous loop within depression and anxiety. In individuals with DSP,

there were several significant patterns in the temporal fluctuations of emotions. Feelings of

depression led to a decrease of positive emotions (depressed -> happy, -0.074, p = 0.032 and

depressed -> content -0.100, p = 0.004) and increase of irritation (0.09, p = 0.006) and anxiety

(0.08, p =0.005) in the next measurement point. Irritation led both to a more positive mood (irritated

-> happy, 0.088, p = 0.001, irritated -> content 0.057, p = 0.030) and less sleepiness (-0.080, p =

0.009) and also to a less anxious state of mind (-0.07, p = 0.003). None of the emotional states were

predicted by sleepiness (p-values of fixed effects coefficient from sleepiness to emotional states all

non-significant. Being anxious predicted sleepiness (0.08, p = 0.04) and happiness (-0.09, p =

0.005). Only depressiveness and anxiousness predicted themselves in the consecutive measurement

points.

In the temporal network, irritation had the highest on in-strength in those without DSP

and happiness in those with DSP, which shows that these emotions were most strongly predicted by

other emotions in these groups. In those without depressiveness and in those with DSP irritation and

depressiveness had high out-strength, meaning that their outgoing temporal associations were

relatively high (compared to other emotions in the network). However, considering the small

number of associations (edge weights) in those without DSP, it appears that none of the emotions

are strong predictors of other emotions.

Contemporaneous networks

The contemporaneous network, including all separate measurement points of each variable,

investigates concurrent relations between emotions. These relations are controlled for temporal

effects and all other emotions the same timepoint to compute a partial (independent of other

associations) correlations network and thus takes the temporal relationship into account. The

relationships within a measurement point can be separately analyzed from relationships between

timepoints in temporal models. In other words, it shows how emotions tend to co-occur at the same

moment, controlling for all other emotions at the same moment and for all temporal relations

among emotions.

Figure 3 shows the contemporaneous network between sleepiness and emotions in

individuals with and without DSP. The contemporaneous networks showed a clustering of positive

and negative emotions in both groups. Specifically, both sleepiness and irritation decreased the

experience of positive emotions concurrently. The contemporaneous networks were strikingly

similar in both groups. However, there were also subtle differences. In individuals with DSP,

sleepiness was associated only with decreased positive emotions, whereas in individuals without

DSP, sleepiness was in inverted association with irritation and in positive association with

depression. In both groups, positive emotions and depressiveness had the highest strength centrality.

Between subject network

Similarly, the between-individual networks (Figure 4) shows the pair-wise associations among the

mean levels of emotions over all measurement points, when adjusting for the mean levels of all

other emotions in the network. The between-subjects model is comparable to cross-sectional

analyses. In general, a positive connection between depressiveness, irritation and anxiety in this

analysis indicate that people with higher scores on depressiveness during a week tend to have

higher scores on anxiety/irritation during the same period and vice versa. In both DSP groups,

positive and negative emotions were again associated with each other and in that sense the results

presented in Figure 4 were relatively similar to those of the contemporaneous networks. In those

without DSP, happiness was negatively associated with irritation but in those with DSP, happiness

was negatively associated with depressiveness. Sleepiness was negatively associated sleepiness in

DSP whereas in non-DSP group sleepiness was associated with irritation. In those without DSP the

most central emotion was irritation, but in those with DSP it was happiness.

DISCUSSION

Poor sleep patterns, sleepiness, and related emotional states form a complex network

of associations that fluctuate over time. As far we know, no studies thus far have focused on

delayed sleep phase and investigated how it affects everyday emotions. In our study, we applied

mlVAR models to analyze the temporal and contemporaneous associations between sleepiness and

multiple emotional states and analyzed whether these networks were different between those

suffering from very late sleep rhythm (DSP) and others. Previous research suggests that DSP is

associated with elevated anxiety and depression scores (Saxvig, Pallesen et al. 2012) and other

problems relating to emotion regulation as well as mental health problems (Sivertsen, Harvey et al.

2015). The causes behind DSP are multiple: biological mechanisms and those relating to

psychological wellbeing are prominent in delaying sleep phase, especially during adolescence

(Micic, Lovato et al. 2016). The same pathways that might lead to DSP are likely to result in poorer

emotional welfare because anxiety and rumination tend to result in later sleep onset times—these

phenomena may also occur reciprocally with late sleep timing (Hiller, Lovato et al. 2014, Stewart,

Gibb et al. 2018). On the other hand, a lack of self-regulating skills is likely to be associated with

later sleep timing or delayed sleep phase (Owens, Dearth-Wesley et al. 2016). These overall

associations have been reported in several studies (Gradisar and Crowley 2013). However, it is not

known how these general challenges are related to dynamic fluctuations of emotions during the day.

In our study, individuals with DSP reported higher average levels of irritation,

depressiveness, and anxiousness and lower levels of contentedness and happiness than those

without. This observation is in line with findings from a large American study investigating over

10,000 adolescents aged 13–18 years. They reported increased risks of mental problems, suicidality,

and mood disorders in relation to greater bedtime delay, and they also found that adolescent sleep

behavior deviated substantially from recommended optimal standards, which was also the case in

our current study (Zhang, Paksarian et al. 2017). We found that, specifically in individuals with

DSP, there were several significant patterns in the temporal fluctuations of emotions, whereas in

individuals without DSP, the only temporal associations were seen in the bidirectional loop between

depression and irritation and in the continuity of anxiety. In the DSP group, feeling depressed led

equally to irritation and anxiety but also to feeling less happy and content at the following

measurement point. Interestingly, in the DSP group, feelings of irritation seemed to be a driving

node that led to feeling less sleepy, happier, and more content and equally less anxious. While

depression preceded irritation, it could show how irritation may function as a catalyst emotion to

feel better.

Without negative feedback loops between emotions unbalanced states that are never

able to recover or reach any positive state of mind would result. Thus, at some point, there need to

be a recovery from negative emotion and in DSP adolescents, and feeling irritated was such as

node. Irritation is not only a negative emotion, but the arousal associated with this emotion may

also potentiate self-directed action to solve the situation causing irritation and then lead to recovery

of the emotion. For adolescents, the irritation maybe also a more frequent emotional response

compared to adults, and, according to the current findings, an important trigger to get out of the

negative emotion loop.

Unexpectedly, sleepiness was not a driving state of mind for specific emotions in the

next time point in either the DSP or non-DSP group; thus, none of the emotional states were

predicted by sleepiness in the previous time point. The DSP group did not report more daytime

sleepiness than the non-DSP group. In objective measurements, they had a later sleep rhythm of 1

hour and 40 minutes, but the difference in sleep duration was only 45 minutes. The mean sleep

duration in the DSP group was over 7 hours during the measurement period, suggesting a sub-

optimal but still reasonable amount of sleep. Previous studies have shown that partial daily sleep

restriction of 2–4 hours increases subjective sleepiness (Lo, Groeger et al. 2012, Slobodanka, Basta

et al. 2013). It was recently reported that restricting sleep for 1 hour over 6 days caused increased

sleepiness (Santisteban, Brown et al. 2019). In that study, however, a similar impact was observed

in the control group with a placebo treatment (sham restriction), which suggests subjective

sleepiness to be affected by the individual’s conception of obtained sleep. Accordingly, the DSP

group in our study may regard their conventional sleep amount sufficient.

Nevertheless, while sleepiness had no association with any of the emotions in

temporal networks, it was positioned as expected in the contemporaneous network. It showed how

feeling sleepy associated with being less happy and content; thus, sleepiness was associated with

reduced positive emotion. Only in the non-DSP group, sleepiness associated also with increased

depression. In both the DSP and non-DSP groups, the different negative and positive feelings

appeared as separate clusters that were inversely correlated.

Although a large body of evidence suggests that poor sleep increases the experience of

negative emotions, reduces the occurrence of positive emotions, and changes the ways in which

individuals experience, understand, express, and regulate these emotions (Walker and van der Helm

2009, Kahn, Sheppes et al. 2013), only a few studies have been able to model and analyze the

dynamic, ongoing process of emotion generation. The strength of the network analyses and

specifically the mlVAR models is the ability to model such complex and potentially causal

relationships. The finding that irritation, rather than depression or anxiety, is the key emotion in

long-term emotional generation and regulation is to be replicated in future studies. This result

would be in line with a body of research showing an association between sleep deprivation and

increased anger and aggression, which was found in men and women and across various age groups

(Saghir, Syeda et al. 2018).

Similar findings were obtained by the previous study using EMA in those with and

without psychiatric disorders. Cousins and co-workers (2011) found that although the relationships

between individual emotions and sleep duration were strongly dependent on present psychiatric

disorders, more time asleep was associated with more positive affect for all diagnostic groups the

following day. However, in adolescents with diagnosed depression disorders, positive emotions

were associated with longer sleep duration in the following night. Higher daytime positive affect in

those with anxiety disorders was associated with less time in bed (Cousins, Whalen et al. 2011).

The findings from the contemporaneous network models as compared to the ones

from the temporal network models suggest that the emotions that we normally think follow each

other, in fact, does that very rapidly or basically are present simultaneously. That means that

positive emotions co-occur with other positive emotions and negative with other negative emotions.

These rapid predictions or co-occurrences followed very similar patterns in both DSP groups.

Fluctuations from positive to negative and from negative to positive happens with a time lag and

that fluctuation seems to be more apparent in those with DSP.

Limitations

There are some limitations that need to be considered. First, the definition of DSP in

our current study was based only on one out of three diagnostic criteria of the full disorder (DSM-

5). The very late sleep rhythm was also assessed only once, without knowing whether it referred to

persistently late or transiently late sleep rhythm. False classifications are possible regarding

subclinical DSP. Nevertheless, the sleep rhythm was defined with actigraphy, which increases

reliability of the assessment. Although EMA has multiple strengths compared to surveys and other

measurement techniques, the intensive nature of EMA studies makes it very difficult to scale up the

number of participants. The participants also must be trained in the use of devices and programs.

However, the target population of this study was probably very familiar with both phones and the

types of devices used. Given the considerable respondent burden involved, there was a risk of

selective drop out (those with DSP sleepiness, depression, or anxiety would not respond). In our

study, the mean number of responses was larger in those with DSP than those without. However, in

general, it would have been preferable to be able to gather more observations per participant from

the mlVAR modeling. One clear advantage of the temporal network over conventional statistical

models is that associations are bidirectional meaning that all variables are simultaneously predictors

and outcomes which provides information about the temporal multivariate relations in the data.

There were more girls than boys in our sample but larger share of boys belonged to

the DSP group. Thus, although the potential effects of gender differences on the association

between emotions is probably small, that needs to be confirmed in the future studies. Our data were

mostly collected during the school year, and data collection was paused for the summer holidays.

The actigraphy measurement period was aimed to represent typical sleep behavior, and we found no

differences between DSP groups across the representativeness of all seven weekdays. We can’t rule

out bias due to residual confounding as this was an observational study. One factor potentially

causing such bias, is socioeconomic status of the family, that may associate with both DSP and

emotions. However, there were no differences in parental socioeconomic status between DSP and

non-DSP groups in our sample and thus that could have technically not been a confounder in the

models.

Conclusions

A delayed sleep phase is extremely common in adolescents but rarely studied beyond

sleep behavior or sporadic health effects. Our study explores practical implications that emerge in

everyday life. The findings in our study highlight that DSP is related to higher levels of irritation,

depressiveness, and anxiousness and lower levels of contentedness and happiness than those with

earlier sleep phases. This suggests that DSP may play a part in every reaction and every emotion a

teenager experiences and thus further emphasizes the need for support for those with DSP.

REFERENCES

Borsboom, D. (2017). "A network theory of mental disorders." World Psychiatry 16: 5-13. Caldwell, J. A., J. L. Caldwell, J. K. Smith and D. L. Brown (2004). "Modafinil's effects on simulator performance and mood in pilots during 37 h without sleep." Aviat Space Environ Med 75(9): 777-784. Cousins, J. C., D. J. Whalen, R. E. Dahl, E. E. Forbes, T. M. Olino, N. D. Ryan and J. S. Silk (2011). "The bidirectional association between daytime affect and nighttime sleep in youth with anxiety and depression." J Pediatr Psychol 36(9): 969-979. Danielsson, K., A. Markstrom, J. E. Broman, L. von Knorring and M. Jansson-Frojmark (2016). "Delayed sleep phase disorder in a Swedish cohort of adolescents and young adults: Prevalence and associated factors." Chronobiol Int 33(10): 1331-1339. Diener, E. and R. A. Emmons (1984). "The independence of positive and negative affect." Journal of Personality and Social Psychology 47: 1105-1117. Epskamp, S. (2015). "bootnet: Bootstrap methods for various network estimation routines." R-package. Epskamp, S., A. O. Cramer, L. J. Waldorp and e. al. (2012). "qgraph: network visualizations of relationships in psychometric data. ." J Stat Softw 12: 1-18. Epskamp, S., L. J. Waldorp, R. Mottus and D. Borsboom (2018). "The Gaussian Graphical Model in Cross-Sectional and Time-Series Data." Multivariate Behav Res: 1-28. Fried, E. I., C. D. van Borkulo, A. O. Cramer, L. Boschloo, R. A. Schoevers and D. Borsboom (2017). "Mental disorders as networks of problems: a review of recent insights." Soc Psychiatry Psychiatr Epidemiol 52(1): 1-10. Gradisar, M. and S. J. Crowley (2013). "Delayed sleep phase disorder in youth." Curr Opin Psychiatry 26(6): 580-585. Hallensleben, N., L. Spangenberg, T. Forkmann, D. Rath, U. Hegerl, A. Kersting, T. W. Kallert and H. Glaesmer (2017). "Temporal Dynamics of Suicide Thoughts. First Results of an Ecological Momentary Assessment Study in Inpatients with depressive Disorders." Zeitschrift Fur Psychosomatische Medizin Und Psychotherapie 63(1): 60-61. Hiller, R. M., N. Lovato, M. Gradisar, M. Oliver and A. Slater (2014). "Trying to fall asleep while catastrophising: what sleep-disordered adolescents think and feel." Sleep Med 15(1): 96-103. Jongeneel, A., G. Aalbers, I. Bell, E. I. Fried, P. Delespaul, H. Riper, M. van der Gaag and D. van den Berg (2020). "A time-series network approach to auditory verbal hallucinations: Examining dynamic interactions using experience sampling methodology." Schizophr Res 215: 148-156. Kahn, M., G. Sheppes and A. Sadeh (2013). "Sleep and emotions: bidirectional links and underlying mechanisms." Int J Psychophysiol 89(2): 218-228. Kahn-Greene, E. T., D. B. Killgore, G. H. Kamimori, T. J. Balkin and W. D. Killgore (2007). "The effects of sleep deprivation on symptoms of psychopathology in healthy adults." Sleep Med 8(3): 215-221. Kolla, B. P., J. P. He, M. P. Mansukhani, S. Kotagal, M. A. Frye and K. R. Merikangas (2019). "Prevalence and Correlates of Hypersomnolence Symptoms in US Teens." J Am Acad Child Adolesc Psychiatry 58(7): 712-720. Kratz, A. L., S. L. Murphy and T. J. Braley (2017). "Ecological Momentary Assessment of Pain, Fatigue, Depressive, and Cognitive Symptoms Reveals Significant Daily Variability in Multiple Sclerosis." Arch Phys Med Rehabil 98(11): 2142-2150.

Lo, J. C., J. A. Groeger, N. Santhi, E. L. Arbon, A. S. Lazar, S. Hasan, M. von Schantz, S. N. Archer and D. J. Dijk (2012). "Effects of partial and acute total sleep deprivation on performance across cognitive domains, individuals and circadian phase." PLoS One 7(9): e45987. Mauss, I. B., A. S. Troy and M. K. LeBourgeois (2013). "Poorer sleep quality is associated with lower emotion-regulation ability in a laboratory paradigm." Cogn Emot 27(3): 567-576. Meyer, C., G. J. Ferrari Junior, R. D. Andrade, D. G. Barbosa, R. C. da Silva, A. Pelegrini and E. P. Gomes Felden (2019). "Factors associated with excessive daytime sleepiness among Brazilian adolescents." Chronobiol Int 36(9): 1240-1248. Micic, G., N. Lovato, M. Gradisar, S. A. Ferguson, H. J. Burgess and L. C. Lack (2016). "The etiology of delayed sleep phase disorder." Sleep Med Rev 27: 29-38. Owens, J., G. Adolescent Sleep Working and A. Committee on (2014). "Insufficient sleep in adolescents and young adults: an update on causes and consequences." Pediatrics 134(3): e921-932. Owens, J. A., T. Dearth-Wesley, D. Lewin, G. Gioia and R. C. Whitaker (2016). "Self-Regulation and Sleep Duration, Sleepiness, and Chronotype in Adolescents." Pediatrics 138(6). Paterson, J. L., J. Dorrian, S. A. Ferguson, S. M. Jay, N. Lamond, P. J. Murphy, S. S. Campbell and D. Dawson (2011). "Changes in structural aspects of mood during 39-66 h of sleep loss using matched controls." Appl Ergon 42(2): 196-201. Powell, D. J. H., C. Liossi, W. Schlotz and R. Moss-Morris (2017). "Tracking daily fatigue fluctuations in multiple sclerosis: ecological momentary assessment provides unique insights." J Behav Med 40(5): 772-783. Rath, D., D. de Beurs, N. Hallensleben, L. Spangenberg, H. Glaesmer and T. Forkmann (2019). "Modelling suicide ideation from beep to beep: Application of network analysis to ecological momentary assessment data." Internet Interv 18: 100292. Saghir, Z., J. N. Syeda, A. S. Muhammad and T. H. Balla Abdalla (2018). "The Amygdala, Sleep Debt, Sleep Deprivation, and the Emotion of Anger: A Possible Connection?" Cureus 10(7): e2912. Santisteban, J. A., T. G. Brown, M. C. Ouimet and R. Gruber (2019). "Cumulative mild partial sleep deprivation negatively impacts working memory capacity but not sustained attention, response inhibition, or decision making: a randomized controlled trial." Sleep Health 5(1): 101-108. Saxvig, I. W., S. Pallesen, A. Wilhelmsen-Langeland, H. Molde and B. Bjorvatn (2012). "Prevalence and correlates of delayed sleep phase in high school students." Sleep Med 13(2): 193-199. Shochat, T., M. Cohen-Zion and O. Tzischinsky (2014). "Functional consequences of inadequate sleep in adolescents: a systematic review." Sleep Med Rev 18(1): 75-87. Sivertsen, B., A. G. Harvey, S. Pallesen and M. Hysing (2015). "Mental health problems in adolescents with delayed sleep phase: results from a large population-based study in Norway." J Sleep Res 24(1): 11-18. Slobodanka, P., M. Basta, A. Vgontzas, I. Kritikou, M. Shaffer, M. Tsaoussoglou, D. Stiffler, Z. Stefanakis, E. Bixler and G. P. Chrousos (2013). "Effects of recovery sleep after one work week of mild sleep restriction on interleukin-6 and cortisol secretion and daytime sleepiness and performance." Am J Physiol Endocrinol Metab( 305): E890–E896,. Stewart, E., B. Gibb, G. Strauss and M. Coles (2018). "Disruptions in the Amount and Timing of Sleep and Repetitive Negative Thinking in Adolescents." Behav Sleep Med: 1-9. Walker, M. P. and E. van der Helm (2009). "Overnight therapy? The role of sleep in emotional brain processing." Psychol Bull 135(5): 731-748. Wang, Z. Y., Z. Z. Liu, C. X. Jia and X. Liu (2019). "Age at menarche, menstrual problems, and daytime sleepiness in Chinese adolescent girls." Sleep 42(6).

Watling, J., B. Pawlik, K. Scott, S. Booth and M. A. Short (2017). "Sleep Loss and Affective Functioning: More Than Just Mood." Behav Sleep Med 15(5): 394-409. Watson, D., L. A. Clark and A. Tellegen (1988). "Development and validation of brief measures of positive and negative affect: The PANAS scales." Journal of Personality and Social Psychology 54: 1063-1070. Zhang, J., D. Paksarian, F. Lamers, I. B. Hickie, J. He and K. R. Merikangas (2017). "Sleep Patterns and Mental Health Correlates in US Adolescents." J Pediatr 182: 137-143.

Figure 1. The distributions of sleepiness and emotional states in those with and without DSP

Figure 2. Temporal networks of sleepiness and emotional states according to DSP status. The green

lines represent positive and red negative partial correlations. Strength centrality represent the extent

how well a node is connected to other nodes. A high in-strength means the node is strongly

predicted by other nodes and high out-strength means that the node is a strong predictor of other

nodes.

Figure 3. Contemporaneous networks of sleepiness and emotional states according to DSP status.

The green lines represent positive and red negative partial correlations. Strength centrality represent

the extent how well a node is connected to other nodes.

Figure 4. Between – subject networks of sleepiness and emotional states according to DSP

status. The green lines represent positive and red negative partial correlations. Strength centrality

represent the extent how well a node is connected to other nodes.

Table 1. Emotional states and number of measurements by groups of DSP

Means and (SD)

No

N=1,833

Yes

N=2,293 p-value

Sleepiness during the day (range 1–8) 4.52 (1.99) 4.56 (1.91) 0.505

Feeling happy (range 1–7) 4.59 (1.37) 4.43 (1.42) 0.001

Feeling content (range 1–7) 4.62 (1.48) 4.40 (1.48) <0.001

Feeling depressed (range 1–7) 1.92 (1.29) 2.19 (1.56) <0.001

Feeling irritated (range 1–7) 2.29 (1.56) 2.42 (1.59) 0.012

Feeling anxious (range 1–7) 2.19 (1.52) 2.39 (1.70) <0.001

Number of measurement times 11.3 (9.55) 11.9 (10.0) 0.030